After a month switching between Claude Code and its open-source rival, I only open Claude for one kind of task

Published Oct 3, 2026, 4:00 PM EDT Nolen began their writing career in 2019, with three years dedicated to editing the Creative section at MakeUseOf. Their expertise lies at the crossroads of technology and creativity, covering areas like photography, video editing, and graphic design. Outside of work, you'll often find Nolen diving into a good book, writing their own stories, or playing video games. Most people reach for Claude Code first, and for good reason, since it's the dominant agent. The catch is the usage limits. Every paid plan draws from a single pool shared by Claude chat, Claude Code, and Claude Design, so going hard in one of them chips away at the others. Once that pool's empty, you either sit tight until the reset or pay standard API rates on top of what you're already paying for the subscription. This is why many are flocking to cheaper alternatives, one of them being OpenCode. It's an open-source coding agent that isn't tied to any one AI company, so you can plug in whatever model you like, including a handful of free ones. After a month of bouncing between them, things did eventually sort themselves into a pattern for my workflow. Want to stay in the loop with the latest in AI? The XDA AI Insider newsletter drops weekly with deep dives, tool recommendations, and hands-on coverage you won't find anywhere else on the site. Subscribe by modifying your newsletter preferences! What OpenCode is actually good for It's built for trying things without counting prompts I first tried OpenCode out of curiosity. Then it turned into my fallback for whenever I didn't feel like burning through my Claude usage. These days it has a real spot in my stack, and it handles a lot more than I expected, mostly thanks to the way it deals with models. The model picker is wild, with 75+ providers to choose from, and since every provider you connect sits in one list, you can swap mid-session. Big Pickle and the other free models are fine for chores, and a paid model only becomes necessary when the job actually calls for one. So for the small stuff, like having a script explained or checking a small component, it can all be handled for free, and spending a paid prompt on that stuff almost feels like a waste now. The fact that it's a terminal tool was my original hang-up, but I've been made aware that there is a desktop app now, which is still in beta. It also reads Claude-style skills from the same .claude/skills folders (with some caveats, which I'll get into) so anything you've built for Claude Code doesn't have to completely stay stuck there. The one job Claude Code still gets Everything else can be an experiment, this can't The one task that sends me back to Claude Code is anything on the systems I actually live in. My Obsidian vault is the big one, along with the /index/ project manager I built inside it, but my vibe-coded projects and the artifacts I use every day end up in the same pile. Those don't get to be experiments since I actually use them every day. They have to work the first time because if something goes wrong I'm the one dealing with the fallout. The /index/ skill is a good example. OpenCode can load it, technically, but the free models I tried would follow it a bit loosely and drift away from the format I'd set up. Part of that is the skill file itself. Claude Code lets a skill carry extra settings at the top, like making it run only when I type /index or limiting which tools it can touch, and OpenCode only reads a handful of basic fields and ignores the rest. Those settings are what keep /index/ from wandering around my vault, so in OpenCode it's running without them. Then there's undo, which was a pretty big oversight on my part when I first thought I could just bring over my whole workflow into OpenCode. OpenCode has undo, and it rolls back file changes, but only for files inside a Git repository (a version-history system you add to a folder). My vault isn't one, so OpenCode's undo doesn't cover it. Claude Code's checkpoints don't need Git - it saves a file's state before each edit and I can rewind from there, although changes made through bash commands aren't tracked, so it's not 100% bulletproof. So it's not one single feature that pulls me back into Claude. It's these small things that add up: the models, skills, and undo. And I'd rather hand off my existing tasks to a system I've already built them in and trust. What each one actually costs me Neither side is free, including the free one Claude is ultimately a subscription, and I pay for it because of the entire ecosystem - Claude Design, the artifacts, projects, the Claude Code GUI with the built-in browser, and the workspace features. Large and long projects here tend to push me into usage credits, which are billed at the standard API rate on top of the subscription. And again, since the limits are shared across the Claude surfaces, a heavy day in one of them eats into the rest. This is honestly the biggest limiting factor of Claude, and I don't really have much else to complain about. OpenCode is free but it still costs me, just in a different way. The free models are capable but thinner than Claude's, there's no question about it. They're great for planning and experimentation, but I probably wouldn't let one loose on my vault. Anthropic also blocked Claude subscription logins in third-party tools earlier this year, and OpenCode pulled its Claude login after a legal request, so to keep using Claude Design and artifacts, I have to keep the subscription regardless of where I can access Claude models. Free covers a lot, just not everything OpenCode turned into the place where I can give things a spin, and only a few weeks in I noticed that I stopped handing stuff to Claude that I wouldn't be upset to lose. I think if OpenCode's undo ever works without Git, I might consider a switch, though the free models would need to stop drifting from my /index/ skill first, and that feels further off. OpenCode